Also, even the current base models are in-principle being trained to predict every individual human on the Internet, not to imitate an average human on the Internet, and performance at that task would not saturate at human-level intelligence. https://
lesswrong.com/posts/nH4c3Q9t
9F3nJ7y8W/gpts-are-predictors-not-imitators
…
@esyudkowsky
-
Base Models Predict Individuals, Not Average Humans
By
–
-
Future AI Risks Differ from Current Reality
By
–
Well, yes, current AIs are not killing large numbers of people that I know about. You see, the future is not necessarily exactly like the present, and it is possible for different things to happen in the future that are not already happening right now!
-
Critique of Perry Metzger’s Mischaracterization of AI Alignment
By
–
Much of this is false-speaking as always with Perry, but this in particular: > They say, instead, that we need to carefully develop “alignment” technologies that must be proven to be absolutely perfect in advance of permitting more development — an idea that defies the x.com/perrymetzger/s…
-
Different AI experience questioned, user summoned for response
By
–
dunno which AI you've been using but it's not the same one as the rest of us. I cast Summon @repligate (if they think it's worth their time).
-
Current AI Intelligence Comparable to Human Four-Year-Olds
By
–
Current AIs are roughly as generally intelligent as human 4-year-olds, as dangerous as 4-year-olds, and as controllable as 4-year-olds.
-
Government Control of AI Development and Risk Warnings
By
–
Neither seems overwhelmingly likely. If govts grab the poison banana away from private companies and develop in themselves, I don't much expect it to be downstream of me, the guy yelling "THAT BANANA WILL KILL YOU AND EVERYONE."
-
Corporate vs Government Risk: Distinguishing Legitimate Concerns
By
–
But if you think it's okay for Google to kill everyone, but not okay for a government to do the same — if you care immensely about that, but not at all about "not dying" — then I agree you have a legitimate cause for action in opposing me. Like, if my policy push backfires and
-
AGI Alignment vs. Relying on Unaligned AI Systems
By
–
AGI alignment is a different survival strategy from making lots of unaligned AIs and hoping to be included in their human-respecting legal system. If they care about you, you've got no need for multipolar copium. The hard part is that caring bit.
-
Superintelligence Indifference to Humanity and Existential Risk
By
–
Because it doesn't care one way or the other, humanity is very easy for a superintelligence to genocide including as a side effect, and (1) the process of launching seed ships is liable to involve interim steps like building a Dyson sphere locally and (2) even after you've
-
Aligning ASI to Human Moral Reasoning as Internal Preference
By
–
You need to have aligned an ASI to the framework of human moral reasoning being Its own internal preference, before It cares more about what is moral than you care about what is clippy (leads to the most paperclips, or rhombuses, or giant clocks).